Silent Evidence

Apex evolves a group of AI agents on a task, so a person doesn't have to tune them by hand. Each agent starts from an idea and writes its own code in a sandbox. After each cycle, a hidden reviewer studies the work and two agents survive. A Mutation step diagnoses what failed and drafts new agents from two parents, and an orchestrator approves or rejects each draft with reasons. A cleaner then erases dates, prices and asset names from the agents' folders, so survivors get no edge in the next cycle. The core knows nothing about the task, which plugs in through a connector. My test task: manage 13 disguised assets through 2025, one week at a time, and beat equal-weight buy-and-hold. In the demo run, 18 agents traded over 4 cycles in about 2 hours, for about $0.71, with no failed API calls. T